“Detecting particles in Cryo-EM micrographs using learned features”
Satya Mallick – Electrical & Computer Engineering, UCSD
In this talk we consider the techniques currently used in the field of object detection and recognition, which can potentially be applied to the problem of particle picking. There are some obvious similarities between the two problems, which encourage us to look into the problem of particle picking from a computer vision viewpoint. On the other hand, there are also some obvious differences, which hint at scope for research and engineering before we can start applying the lessons learned from the computer vision and machine learning community. Particle picking can be seen as yet another object detection problem, which could be solved using a generative or discriminative learning algorithm which is trained using a large number of example images of particles (and possibly non-particles). We consider a well researched area in computer vision, the problem of detecting faces in a given image. The challenge in face detection arises because there is large within-class variation in the images of faces due to difference amongst individuals as well as pose, scale, and lighting variation. In contrast, in particle picking the within-class variation is small, and the problem of scale and lighting are minor. Yet, automated cryo-EM particle picking must confront very low signal to noise ratios (SNR) whereas the noise in video images is a minor nuisance. We derive hope from the fact that feature-based face detection algorithms are sufficiently accurate and have already achieved real-time performance. Finally, we report on the results of our algorithm which is inspired by the current state of the art in face detection and compare the results with other particle detection algorithms.